Clarke versus Peters - robopoll results

Disappointed Peters supporter

I've called the Peters campaign phone on two weekends, and had no answer. As far as I could tell, they did no Memorial Day parades. And I visited the listed HQ last Sunday, but it was closed and empty looking.

This is a far cry from 2008 and 2010, where the campaign was already hopping by this time (for a General election, let alone a primary)!

---

by: William Allen Simpson @ Fri Jun 15, 2012 at 22:37:51 PM UTC

Wait...the pool was only 168 respondents??

Doesn't that make the margin of error almost +/- 8%?

If I understand MoE correctly (and I'm sure you'll correct me if I don't), that means that it could range anywhere from Clarke 41 / Peters 41 to Clarke 57 / Peters 25, yes?

---

by: Brainwrap @ Sat Jun 16, 2012 at 01:12:54 AM UTC

Margin of error of one candidate's lead over another

Not exactly.  What we call "margin of error" applies to each marginal, but not to both at the same time.  

To calculate (say) a 95% confidence interval on the DIFFERENCE of two marginals requires estimating their covariance, which is less than 1.00 because there are other candidates in the race.  In simpler terms, if we assume the poll has overestimated Clarke's percentage by 8%, we would only expect it to underestimate Peters's margin by 33%/52% * 8%, or about 5%.  I haven't calculated it rigorously, but I think the 95% two-tailed confidence interval on the difference is about 13%.

My preferred method is to look at the log-odds:

LN(82/55) = 0.399

The standard error of the log-odds is closely approximated by

SQRT(1/82 + 1/55) = 0.174

So Clarke's lead is 2.29 standard errors, which is well past the two-tailed 95% confidence interval that everybody likes so much.  This method eliminates the need to think about covariance at all - it's wonderfully robust.

And thank you for asking!

---

by: Grebner @ Sat Jun 16, 2012 at 02:29:53 AM UTC

[ Parent ]

undecided?

why was there no undecided option?  also, is it normal to skip the weekend? Does skipping the weekend have a known effect one way or the other for getting responses?

---

by: ibelieveinMichigan @ Sat Jun 16, 2012 at 02:20:18 AM UTC

Great questions!

1) There are no undecideds because PPC's standard method is not to offer the choice.  Other pollsters disagrees with us on this point, but we've decided that offering undecided as an option tends to reduce the number of useful responses we get ("useful" being defined as somebody actually choosing a candidate) while we've never found anything useful in analyzing the undecided votes.

Our standard question is roughly:  "When you vote in the Democratic primary in August, the candidates for Congress will be Hanson Clarke, .....  Press 1 if you will vote for Hanson Clarke.  Press 2 if you..."

Our results tell us who strongest among the candidates.  We assess how many people are paying attention to the race, and are knowledgeable enough to choose a candidate, by looking at the fraction who participate.  That analysis is reflected in the first paragraph of the press release reprinted above.

2) We don't like to poll on Fridays or Saturdays, because we don't get much useful data for our money.  We poll either Sunday afternoon or Sunday evening, depending on our whim, as well as Monday through Thursday evenings.  I suppose there may be some theoretical bias to this choice, but we haven't seen anything specific.  Some robo-pollers run each poll on a single day, which we reject for aesthetic reasons.

Various accidents have allowed us to examine a number of twists, none of which seem to affect our results.  Once we called at 7AM Sunday, rather than 7PM - we didn't get many completed interviews, but the pattern was identical to what we got the next night.  Several times, we've butchered the pronunciation of a name, but that didn't seem to bother the voters.  On several occasions we've called the WRONG DISTRICT, and the result was truly surprising:  90% fewer responses than when we repeated the survey later in the correct district - people actually figured out that they had no idea who the presented candidates were.  I found that truly encouraging, not only about our methods but also about the voting public.

I always describe PPC's polling as "cheap and second-rate".  If one of our shortcuts introduces a small random bias, I figure it won't really matter because our results aren't all that precise anyway.

We do pretty well, and our results are generally right.  In the present case, please note the absolute silence from the other candidates.

---

by: Grebner @ Sat Jun 16, 2012 at 02:53:13 AM UTC

[ Parent ]

Silence

Since the Lawrence Campaign requested the courtesy of a phone call fro
you the day after Mr. Clarke released the poll and we continue to wait for the call, it is a little odd for you to comment on our silence.  I think you have our contact info.  It is in your client files.  

---

by: Jensen @ Sat Jun 16, 2012 at 18:14:02 PM UTC

[ Parent ]

First I've heard of it.

It's a little weird to be working out the details of a phone call in a political blog, but here we are.

Was it a general call to PPC?  If so, I bet it got swept up by the sales people who generally intercept people who think they need to talk to me personally, but whom the company considers potential sales prospects.

The best method of reaching me personally is always email: Mark dot Grebner at GMail dot com.  I respond to email almost around the clock.

My comment about silence was really about responses in the media.  If anybody had poll results which conflict with my numbers, I'd have thought they'd release them.  Since we've heard nothing, I infer my numbers are in the right ballpark.

---

by: Grebner @ Sat Jun 16, 2012 at 18:40:38 PM UTC

[ Parent ]